unity-pipeline

Unity Pipeline (live Editor control)

Stack: unity CLI (~/.unity/bin/unity, normally already on PATH) plus the UPM package com.unity.pipeline, pinned in Aspid.FastTools/Packages/manifest.json. The Editor exposes ~460 commands; the CLI just forwards to them, so no CLI update is needed when the package adds one.

Generic CLI surface (editors, licenses, headless build/test, mcp) is covered by the vendor unity-cli skill. This file covers only what is specific to Aspid.FastTools.

1. Always target a project explicitly

Several Editors run at once here — the main checkout plus .claude/worktrees/shared-*. A bare unity command … fails with COMMAND_FAILED / Multiple Unity Editor instances found.

P="$PWD/Aspid.FastTools"            # from the repo (or worktree) root
unity command <name> --project-path "$P" --format json

2. Health check — status, not pipeline list

unity status --format json          # port, project, version, PID, state per Editor

Reads the per-Editor lockfile instead of probing over HTTP: faster, and stale instances are reported as unreachable. unity pipeline list is the wrong tool for this — a closed Editor can still show Running=true there from a leftover lock. Use pipeline list only for package install/version questions.

unity list --project-path "$P" prints every command the Editor exposes with its parameter schema — use it instead of guessing argument names.

3. Edit → verify loop

unity command set_autotick --enable true --project-path "$P"   # do this first
unity command recompile --project-path "$P"
unity command recompile_status --project-path "$P"             # poll: idle|triggered|compiling|completed|up_to_date
unity command run_tests --mode editor --filter <Fixture> --project-path "$P"

• set_autotick first. Unfocused Unity throttles its update loop, so a recompile or test run started from the shell can hang indefinitely without it.
• The server goes silent during domain reload — commands return empty. Poll with retries; the connection restores itself afterwards (no reinstall needed).
• run_tests right after another command can return result=null (runner still busy) — retry.
• Long runs: --async_tests true, then poll test_status; abort with cancel_tests.